After a brief review of machine learning techniques and applications, this Ph.D. thesis examines several approaches for implementing machine learning architectures and algorithms into hardware within our laboratory. From this interdisciplinary background support, we have motivations for novel approaches that we intend to follow as an objective of innovative hardware implementations of dynamically self-reconfigurable logic for enhanced self-adaptive, self-(re)organizing and eventually self-assembling machine learning systems, while developing this new particular area of research. And after reviewing some relevant background of robotic control methods followed by most recent advanced cognitive controllers, this Ph.D. thesis suggests that amongst many well-known ways of designing operational technologies, the design methodologies of those leading-edge high-tech devices such as cognitive chips that may well lead to intelligent machines exhibiting conscious phenomena should crucially be restricted to extremely well defined constraints. Roboticists also need those as specifications to help decide upfront on otherwise infinitely free hardware/software design details. In addition and most importantly, we propose these specifications as methodological guidelines tightly related to ethics and the nowadays well-identified workings of the human body and of its psyche.

Selon des études comportementales, les comportements complexes sont des processus multi-échelles, souvent composés de sous-éléments (unités fonctionnelles ou primitives). Cette thèse propose des architectures fonctionnelles afin de représenter la structure dynamique des unités fonctionnelles ainsi que celle des comportements multi-échelles résultants. Dans un premier temps, des unités fonctionnelles sont modélisées comme des flux structurés de faible dimension dans l'espace de phase (modes de fonctionnement). Des dynamiques supplémen-taires (signaux opérationnels) opèrent sur ces modes de fonctionnement faisant émerger des comportements complexes et sont classifiés selon la séparation entre leur échelle temporelle et celle des modes. Ensuite, des mesures de complexité, appliquées sur des architectures dis-tinctes composant un mouvement simple, révèlent un compromis entre la complexité des modes de fonctionnement et celle des signaux opérationnels. Celui-ci dépend de la séparation entre leurs échelles temporelles et soutient l'efficacité des architectures utilisant des modes non triviaux. Dans un deuxième temps, une architecture pour le comportement séquentiel (ici l'écriture) est construite via le couplage des modes de fonctionnement (réalisant des lettres) et des signaux opérationnels, ceux-ci beaucoup plus lents ou beaucoup plus rapides. Ainsi, l'importance des interactions entre les échelles temporelles pour l'organisation du comporte-ment est illustrée. Enfin, les contributions des modes et des signaux sur la sortie de l'architec-ture sont déterminées. Ceci semble être uniquement possible grâce à l'analyse du flux de phase (c'est-à-dire, non pas à partir des trajectoires dans l'espace de phase ni des séries temporelles). / Behavioural studies suggest that complex behaviours are multiscale processes, which may be composed of elementary ones (units or primitives). Traditional approaches to cognitive mod-elling generally employ reductionistic (mostly static) representations and computations of simplistic dynamics. The thesis proposes functional architectures to capture the dynamical structure of both functional units and the composite multiscale behaviours. First, a mathe-matical formalism of functional units as low dimensional, structured flows in phase space is introduced (functional modes). Second, additional dynamics (operational signals), which act upon functional modes for complex behaviours to emerge, are classified according to the separation between their characteristic time scale and the one of modes. Then, complexity measures are applied to distinct architectures for a simple composite movement and reveal a trade off between the complexities of functional modes and operational signals, depending on their time scale separation (in support of the control effectiveness of architectures employing non trivial modes). Subsequently, an architecture for serial behaviour (along the example of handwriting) is demonstrated, comprising of functional modes implementing characters, and operational signals much slower (establishing a mode competition and ‘binding’ modes into sequences) or much faster (as meaningful perturbations). All components being coupled, the importance of time scale interactions for behavioural organization is illustrated. Finally, the contributions of modes and signals to the output are recovered, appearing to be possible only through analysis of the output phase flow (i.e., not from trajectories in phase space or time).